feat: Add embedded structs based on result match

[kylecarbs]

Identifies every column with a returned struct for queries.

This is a breaking change. I'm very open to comments, ideas, or reasons why this shouldn't even happen. To prevent breakage we could move this under a configuration option, and eventually migrate to default in a major version.

-- schema.sql CREATE TABLE users (id text, email text); CREATE TABLE usernames (user_id text, name text); -- query.sql -- name: GetUsers :many SELECT * FROM users; -- name: GetUsersWithUsername :many SELECT users.*, usernames.name FROM users JOIN usernames ON users.id = usernames.user_id;

Before:

type GetUsersWithUsernameRow struct { ID sql.NullString `db:"id"` Email sql.NullString `db:"email"` Name sql.NullString `db:"name"` }

After:

type GetUsersWithUsernameRow struct { User Name sql.NullString `db:"name"` }

[cmoog]

Run make regen at the root to regenerate the expected output for e2e tests.

[cmoog]

Didn't realize the join handling hadn't reached master, but this will be a problem when #983 lands.

Handling fields that are nullable after joins, but not nullable in the data model will be difficult with this approach.

[kylecarbs]

The embedded struct shouldn't be nullable though... right? It doesn't represent any data itself, simply a container for data.

[cmoog]

Correct. But the problem comes when Foo.A is not null in the data model, but can be null after a left join. When we marshal a possibly nullable field to Foo.A directly this nullability type information is lost. Instead we'd want to use sql.NullString for Foo.A in that particular query. More context here: #983

After those PRs land, adding proper handling for each join case to this PR will be straightforward. It's possible those cases will be automatically taken care of by #983 before this code path is even reached, not quite sure implementation wise.

[kylecarbs]

Ahh understood.

[cmoog]

bump

[kyleconroy]

This PR is now out of date, there are merge conflicts.

Also, I'm not sure how you're planning on solving the LEFT JOIN issue. You can't use the existing structs because a non-nullable column can now be null. I think you'll need to generate new structs, which may defeat the entire purpose. Maybe we just don't allow embedding to work when using left joins?

Lastly, the syntax for this needs to be a special sqlc function. I'd suggest using sqlc.embed(). The reason is that overloading the table.* method isn't backwards compatible.

[kyleconroy]

@kylecarbs I haven't heard from you in a while, so I'm going to close this out. If you end up working on this, please open a new pull request or push to this one and mention me in a comment.

[ludusrusso]

@kylecarbs I'm interesting in this feature. If you are not working on this anymore I can take it!

[kylecarbs]

Feel free! I'd love to tackle it but don't quite have bandwidth at the moment.